batch_iterate and abatch_iterate Finally Agree

What it does

batch_iterate() and its async twin abatch_iterate() are the helpers you reach for when you need to split a big list of inputs into fixed-size chunks before feeding them to a model or a chain. The bug that langchain-core 1.5.5 fixes: when you passed None as the batch size, or a batch size of 0, the async version behaved differently from the sync one. One would raise, the other would silently return nothing - or a zero-size batch could spin out an empty or infinite generator. The fix makes the async path match the sync path exactly: None means "no batching" (yield the whole list as one batch) and 0 is rejected consistently with a clear error. Your code now behaves identically whether you call the sync or async variant.

Why it matters

Real-world LangChain apps routinely chew through large input sets - thousands of documents to embed, hundreds of prompts to run, a big CSV of rows to classify. batch_iterate and abatch_iterate are the idiomatic way to chunk those inputs so you do not blow up memory or trip provider rate limits. The danger of a sync/async mismatch is subtle: you write a pipeline that works fine with batch_iterate, then switch to abatch_iterate for concurrency and suddenly get different chunking - or a crash - for edge-case batch sizes. This fix removes that footgun, so you can confidently use the async variant in production with the same semantics. That matters for any agent or ETL job that must behave predictably at scale.

Example

$ Chunk a list of 10 inputs into batches of 3, then confirm None means "no batching"

batch_iterate(size=3):
   [0, 1, 2]
   [3, 4, 5]
   [6, 7, 8]
   [9]
batch_iterate(size=None):
   [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
abatch_iterate(size=3):
   [0, 1, 2]
   [3, 4, 5]
   [6, 7, 8]
   [9]
batch_iterate(0) -> batch_size must be a positive integer

$ Reject a zero batch size consistently in both variants

try:
    list(batch_iterate(0, inputs))
except ValueError as e:
    print("batch_iterate(0) ->", e)

# batch_iterate(0) -> batch_size must be a positive integer

Common flags

batch_iterate

Sync helper that chunks an iterable into fixed-size batches.

abatch_iterate

Async twin of batch_iterate; now consistent for None and zero size.

Runnable.batch

Batch-invoke a runnable over a list of inputs (built-in batching).

Runnable.abatch

Async batch-invoke a runnable over a list of inputs.